The Necessity Of A Power Systems 911
September 25, 2017 Timothy Prickett Morgan
People are starting to talk about the future, and in some cases perhaps impending, Power9 systems from IBM and its OpenPower partners. The chatter is about machines with two, four, eight, or 16 processors, so far at least. But the one thing I am not hearing anyone talk about is an entry Power box, either with or without the Cognitive Systems label, that has a modestly configured processor complex that is suitable for the IBM i workloads in the current customer base and yet allows them to affordably and quickly expand their capacity to absorb more and more workloads in their datacenters.
We are talking about machines like the AS/400 Model 150 and 270, the M25 and M50 in the Power6 generation, or the four-core variants of the Power7 and Power8 machines, just to name a few of the classic and well-selling entry machines that supported OS/400 and IBM i. These extremely entry machines have been in the P05 software tier, which makes them more affordable than systems in the P10 and P20 tier that tend to have one, two, and sometimes four sockets and that are more expensive on all fronts: CPU, memory, disk, and systems software and tools.
In recent years, IBM has been somewhat reluctant to embrace the low-end of the IBM i customer base. When the Power8 machines initially launched back in April 2014, there were five machines launched and only three of them ran IBM i and none of them had relatively modest configurations by the standard of the time. That included the Power S814 with one socket and the Power S824 with two sockets, both of which came in a 4U form factor. This is pretty fat for a modern server, and these days, you can get an eight socket machine in 4U or 5U of space in a rack.
At announcement time, IBM was talking about possibly putting out a four-core Power8 machine aimed at entry users in September 2014, and after much complaining from the base and wiseguys like us, IBM jumped the gun and actually delivered a variant of the Power S814 with a four-core processor that was capped at 64 GB of main memory and a maximum of eight media drives, be they flash or disk. This machine used 3.02 GHz Power8 cores and was rated at an aggregate of 39,500 CPWs of IBM i performance, and it replaced a Power 720+ that had four cores running at 3.6 GHz and that was rated at 28,400 CPWs across those four cores.
Earlier this year, on Valentine’s Day, IBM pushed the envelope on entry downwards with the Power S812 Mini, which is a version of the Linux-only Power S812L single-socket Power8 machine that has been allowed to run IBM i and that has a single 3.03 GHz core activated and 64 GB of main memory; it would be rated at somewhere just under 10,000 CPWs, and at $16,111 for a configured hardware system and another $19,580 for a full software stack, it was, by IBM i standards at least, reasonably priced. The idea was for the single core Power S812 Mini to be about 20 percent lower cost than a single core activated on a four-core Power S814; it was about 35 percent cheaper than the Power S822 and 60 percent cheaper than the Power S824 with a similar configuration and a single core activated, as we showed.
We have not heard of a Power S912 Mini, but if all the changes IBM has made with the Power9 chip come to pass, then a single core should deliver somewhere around twice the performance of a Power8 core, or around 20,000 CPWs. So, a single core Power9 machine – a single core running on a single socket in a 1U chassis for the sake of fun would be called a Power S911, emergency emergency – would have almost as much oomph as a four-core Power7+ machine from five years ago. And because software is priced per core, a P05 machine with a single Power9 core would be considerably cheaper than a three core Power7+ machine with roughly the same performance. Assuming you can gear the Power9 chip down to 3 GHz or so, of course.
People familiar with IBM’s Power9 server plans indicate that there are multiple two-socket machines in the pipeline, and something analogous to the Power S814 and Power S824 machines and presumably called the Power S914 and Power S924. Depending on the pricing and features, these might be suitable as a Power9 Mini model, but it is my guess that they will be overkill, just like the four-core Power S814 was for a lot of customers. Particularly if IBM goes with a four-core rather than a one or two core processor card.
We appreciate that IBM is focused on the fight for the datacenter with Intel, and now AMD and a few ARM suppliers. We appreciate all of the engineering IBM is doing to create compelling and powerful systems. We just don’t want Big Blue to forget entry IBM i shops. It is not too late to put together an attractive Power9 Mini system that has more capacity, and that allows for the integration of all the new goodies like flash and GPU acceleration, too. If IBM wants the Cognitive Systems business, as Power Systems is now called, to grow, it has to think of those 125,000 or so IBM i shops who don’t need to deploy 24 cores with heavy SMT8 threading across two sockets with 4 TB of main memory. They need to start somewhere to grow into such a big box.
RELATED STORIES
The Bang For The Buck Of Entry IBM i Servers
The Lowdown On Pricing For The Power S812 Mini
IBM i License Transfer Deal Comes To The Power S812 Mini
IBM Gives The Midrange A Valentine’s Day (Processor) Card
More Insight Into The Rumored Power Mini System
Geared Down, Low Cost Power IBM i Box Rumored
IBM Cuts Core And Memory Pricing On Entry Power Iron
Entry Power8 Systems Get Express Pricing, Fat Memory
Reader Feedback On Power S814 Power8 Running IBM i
Four-Core Power8 Box For Entry IBM i Shops Ships Early
I think IBM traditionally tries to supply to those who are already pushing the envelope on the earlier processors. To me this makes sense. The newer lower end Power 8’s are recent editions and are plenty good enough for the small market. After they fill the initial need for higher end Power 9 models then they’ll come out with lower end versions. It may take a year or two, but again, the Power 8 variants won’t be that long in the tooth by then.
Mr. Morgan, you assume that IBM wants the i to be viable long-term. I don’t think that is the case.
I do assume that. Correct!